Re: configuration where 2 out of 3 physical servers can fail

[Mikael Ronström]

When running a cluster between 3 DCs one would normally use
a normal cluster with 2 replicas in each DC and then use
MySQL Replication to set up replication between the DCs.

This works perfectly fine, but is obviously a bit more
work to setup. MySQL Cluster can handle circular replication
for this case and there is even options to handle
conflict detection of one allows updates to any cluster.

Using normal MySQL Cluster with 3 replicas will work to
some extent as mentioned in the previous post.
Ensure that the management server is located on a separate
VM to the data nodes in each DC. In that case you will
survive all single VM failures. You will also survive
a DC failure unless the arbitrator was in the DC that
failed.

I would recommend using MySQL Cluster with MySQL Replication
between the DCs, this also ensures that you don't have to
run a synchronized transactions over a wide area network
that will take about 5-10x the time it takes to send a
message between DCs (1-2 ms if in same region, 20-40 ms
if in different regions).